zfs_host: zed and pool capacity and scrub metrics for the node exporter textfile collector

zfs_host/defaults/main.yml

@@ -0,0 +1,4 @@
+---
+# Directory the node exporter reads textfile metrics from (the Debian default).
+zfs_host_textfile_dir: /var/lib/prometheus/node-exporter
+zfs_host_metrics_interval: 5min

zfs_host/tasks/main.yml

@@ -0,0 +1,41 @@
+---
+# ZFS event daemon and pool metrics for the node exporter. zed keeps its default zed.rc and
+# logs to the journal, which journal_upload ships.
+- name: Install the ZFS event daemon
+  ansible.builtin.apt:
+    name: zfs-zed
+    install_recommends: false
+    state: present
+
+- name: Enable the ZFS event daemon
+  ansible.builtin.systemd:
+    name: zfs-zed
+    enabled: true
+    state: started
+  # Check mode does not install the package.
+  ignore_errors: "{{ ansible_check_mode }}"
+
+- name: Install the pool metrics script
+  ansible.builtin.template:
+    src: zfs-pool-metrics.j2
+    dest: /usr/local/sbin/zfs-pool-metrics
+    mode: "0755"
+
+- name: Install the pool metrics units
+  ansible.builtin.template:
+    src: "{{ item }}.j2"
+    dest: /etc/systemd/system/{{ item }}
+    mode: "0644"
+  loop:
+    - zfs-pool-metrics.service
+    - zfs-pool-metrics.timer
+  register: zfs_host_units
+
+- name: Enable the pool metrics timer
+  ansible.builtin.systemd:
+    name: zfs-pool-metrics.timer
+    enabled: true
+    state: started
+    daemon_reload: "{{ zfs_host_units.changed }}"
+  # Check mode does not write the units.
+  ignore_errors: "{{ ansible_check_mode }}"

zfs_host/templates/zfs-pool-metrics.j2

@@ -0,0 +1,49 @@
+#!/bin/bash
+# Pool capacity and the last scrub for the node exporter textfile collector.
+# A scrub in progress or cancelled keeps the previous end time: the pool has no newer completed scrub.
+set -euo pipefail
+
+dir={{ zfs_host_textfile_dir }}
+out=$dir/zfs_pool.prom
+tmp=$(mktemp "$dir/.zfs_pool.XXXXXX")
+trap 'rm -f "$tmp"' EXIT
+
+# Value of a metric line of the previous run, 0 if there is none.
+previous() {
+  awk -v m="$1" -v p="$2" 'index($0, m "{pool=\"" p "\"}") == 1 {v = $2} END {print v + 0}' "$out" 2>/dev/null || echo 0
+}
+
+status=$(zpool status -p)
+{
+  echo '# TYPE zfs_pool_size_bytes gauge'
+  echo '# TYPE zfs_pool_allocated_bytes gauge'
+  echo '# TYPE zfs_pool_fragmentation_ratio gauge'
+  zpool list -Hp -o name,size,allocated,fragmentation | while IFS=$'\t' read -r pool size alloc frag; do
+    echo "zfs_pool_size_bytes{pool=\"$pool\"} $size"
+    echo "zfs_pool_allocated_bytes{pool=\"$pool\"} $alloc"
+    if [[ $frag =~ ^[0-9]+$ ]]; then
+      echo "zfs_pool_fragmentation_ratio{pool=\"$pool\"} $(awk -v f="$frag" 'BEGIN {print f / 100}')"
+    fi
+  done
+
+  echo '# TYPE zfs_pool_scrub_end_timestamp_seconds gauge'
+  echo '# TYPE zfs_pool_scrub_errors gauge'
+  zpool list -Hp -o name | while read -r pool; do
+    scan=$(awk -v p="$pool" '$1 == "pool:" {cur = $2} cur == p && $1 == "scan:" {sub(/^[ \t]*scan:[ \t]*/, ""); print; exit}' <<< "$status")
+    end=0
+    errors=0
+    if [[ $scan =~ ^scrub\ repaired\ .*\ with\ ([0-9]+)\ errors\ on\ (.+)$ ]]; then
+      errors=${BASH_REMATCH[1]}
+      end=$(date -d "${BASH_REMATCH[2]}" +%s)
+    elif [[ $scan =~ ^scrub\ (in\ progress|canceled) ]]; then
+      end=$(previous zfs_pool_scrub_end_timestamp_seconds "$pool")
+      errors=$(previous zfs_pool_scrub_errors "$pool")
+    fi
+    echo "zfs_pool_scrub_end_timestamp_seconds{pool=\"$pool\"} $end"
+    echo "zfs_pool_scrub_errors{pool=\"$pool\"} $errors"
+  done
+} > "$tmp"
+
+chmod 0644 "$tmp"
+mv "$tmp" "$out"
+trap - EXIT

zfs_host/templates/zfs-pool-metrics.service.j2

@@ -0,0 +1,6 @@
+[Unit]
+Description=ZFS pool metrics for the node exporter textfile collector
+
+[Service]
+Type=oneshot
+ExecStart=/usr/local/sbin/zfs-pool-metrics

zfs_host/templates/zfs-pool-metrics.timer.j2

@@ -0,0 +1,10 @@
+[Unit]
+Description=ZFS pool metrics for the node exporter textfile collector
+
+[Timer]
+OnBootSec=1min
+OnUnitActiveSec={{ zfs_host_metrics_interval }}
+AccuracySec=10s
+
+[Install]
+WantedBy=timers.target